learning_rate: 0.0001
train_batch_size: 16
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 5000
num_epochs: 30
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.2979	0.4877	1000	3.2300	100.0	99.2291
0.7248	0.9754	2000	0.4206	39.0933	10.3811
0.3723	1.4628	3000	0.2100	29.3695	7.3497
0.329	1.9505	4000	0.1888	27.7261	6.8250
0.272	2.4379	5000	0.1598	26.2061	6.2687
0.2413	2.9256	6000	0.1445	25.4042	6.0652
0.1809	3.4131	7000	0.1366	24.2543	5.7535
0.1936	3.9008	8000	0.1321	24.0472	5.6091
0.1516	4.3882	9000	0.1344	23.5053	5.4742
0.1724	4.8759	10000	0.1288	23.2410	5.4553
0.139	5.3633	11000	0.1289	22.8885	5.2951
0.1445	5.8510	12000	0.1243	22.9414	5.2691
0.1201	6.3385	13000	0.1193	22.6329	5.2115
0.1291	6.8261	14000	0.1154	22.1527	5.0631
0.1021	7.3136	15000	0.1118	21.6328	4.9156
0.1081	7.8013	16000	0.1145	21.8575	5.0008
0.1032	8.2887	17000	0.1118	21.4390	4.8651
0.1073	8.7764	18000	0.1098	21.1350	4.7972
0.0872	9.2638	19000	0.1119	21.1438	4.8130
0.0997	9.7515	20000	0.1031	20.8706	4.7120
0.0817	10.2390	21000	0.1086	20.7384	4.6718
0.081	10.7267	22000	0.1050	20.9147	4.6828
0.0674	11.2141	23000	0.1103	20.6371	4.6386
0.0727	11.7018	24000	0.1119	20.6459	4.6284
0.0711	12.1892	25000	0.1059	20.4520	4.5676
0.068	12.6769	26000	0.1027	20.3507	4.4958
0.0654	13.1644	27000	0.1030	20.3375	4.5274
0.064	13.6520	28000	0.1066	20.4564	4.5282
0.0645	14.1395	29000	0.1056	20.3066	4.4974
0.0729	14.6272	30000	0.1016	19.9366	4.4059
0.053	15.1146	31000	0.1083	19.7956	4.3356
0.0534	15.6023	32000	0.1051	19.7779	4.3467
0.058	16.0897	33000	0.1060	19.7471	4.3285
0.0582	16.5774	34000	0.1034	19.7559	4.3104
0.046	17.0649	35000	0.1052	19.7383	4.2938
0.0433	17.5525	36000	0.0995	19.7163	4.2836
0.0503	18.0400	37000	0.1001	19.5488	4.2252
0.0407	18.5277	38000	0.0991	19.5841	4.2378
0.0443	19.0151	39000	0.0979	19.4563	4.1834
0.0443	19.5028	40000	0.1061	19.3374	4.2228
0.0412	19.9905	41000	0.1016	19.2801	4.1849
0.0426	20.4779	42000	0.1036	19.3109	4.2039
0.0332	20.9656	43000	0.1027	19.0598	4.1242
0.0348	21.4531	44000	0.0988	19.1083	4.1210
0.0505	21.9407	45000	0.0978	19.0422	4.0918
0.0397	22.4282	46000	0.1000	18.9452	4.0832
0.0421	22.9159	47000	0.1001	19.0466	4.0879
0.0413	23.4033	48000	0.0988	18.9100	4.0508
0.0316	23.8910	49000	0.1009	18.9805	4.0642
0.0367	24.3784	50000	0.0999	18.8307	4.0303
0.0312	24.8661	51000	0.0982	18.8307	4.0248
0.032	25.3536	52000	0.0991	18.7822	4.0082
0.03	25.8413	53000	0.0988	18.7117	4.0027
0.0336	26.3287	54000	0.0969	18.7558	4.0074
0.0272	26.8164	55000	0.0993	18.6721	3.9916
0.0337	27.3038	56000	0.0994	18.6853	3.9964
0.0361	27.7915	57000	0.0995	18.7029	3.9964
0.0378	28.2790	58000	0.0994	18.6809	3.9845
0.0268	28.7666	59000	0.0989	18.6765	3.9853
0.0364	29.2541	60000	0.0986	18.6897	3.9901
0.0354	29.7418	61000	0.0987	18.6853	3.9908

Framework versions

Transformers 4.56.1
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.22.0

davidilag/wav2vec2-xls-r-300m-pt-500h-FO-500h-IS-cp-best-faroese-100h-30-epochs_run9_2025-09-10

wav2vec2-xls-r-300m-pt-500h-FO-500h-IS-cp14-faroese-100h-30-epochs_run9_2025-09-10

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-pt-500h-FO-500h-IS-cp-best-faroese-100h-30-epochs_run9_2025-09-10